DeepfakeBench-MM: A Comprehensive Benchmark for Multimodal Deepfake Detection

arXiv:2510.22622 · cs.CR, cs.CV, cs.MM · Submitted 2025-10-26 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Security Radio. Generated commentary on the latest security and cryptography papers.

Nadia: We're looking at the paper: "DeepfakeBench-MM: A Comprehensive Benchmark for Multimodal Deepfake Detection".

Elias: Here's the gist of the paper. DeepfakeBench-MM introduces Mega-MMDF, a large-scale and diverse multimodal deepfake dataset and DeepfakeBench-MM, the first unified benchmark for multimodal deepfake detection The paper presents this infrastructure to address the lack of sufficient and diverse training data and standardized benchmarks hindering deeper exploration Dataset Construction Mega-MMDF is a large-scale multimodal deepfake dataset constructed through an automated deepfake generation framework, featuring three key characteristics: large scale, high forgery diversity, and high quality The dataset contains 0.1 million real samples and 1.1 million forged samples, totaling 1.2 million instances The construction involves generating 21 compositionally diverse forgery pipelines based on combinations of 28 forgery methods across 7 forgery types Quality control is integrated into every pipeline through three assessment modules: audio quality assessment, visual quality assessment, and audio-video synchronization quality assessment Overall dataset quality was validated using objective quantitative metrics like Fréchet Audio Distance (FAD) and Fréchet Video Distance (FVD), alongside human studies involving 60 volunteers Benchmark Development DeepfakeBench-MM is the first unified benchmark for multimodal deepfake detection, supporting 5 multimodal deepfake datasets and 11 detectors within an extensible codebase The benchmark standardizes training/evaluation protocols across detection components and provides a common platform for method analysis Supported datasets include FakeAVCeleb, LAV-DF, AVDeepfake1M, IDForge, and Mega-MMDF Detectors implemented span four categories: Baseline Models, Regular Models (7 detectors), Ensemble Models (one model), and Pretrained Multimodal Large Language Models (two models) The evaluation protocol focuses on intra-dataset performance, cross-dataset generalization, and cross-pipeline generalization using the AUC metric Analysis of Findings Analysis 1 observed that although detectors generally achieve high AUCs within Mega-MMDF (up to 0.994), their performance drops markedly when applied to external datasets Analysis 2 revealed that in stacked forgeries, artifacts contribute unequally, following the hierarchy EFS > FS > FR Analysis 3 showed that finetuning the second-stage backbones of detectors like AVTS and AVFF yields consistent improvements across datasets Analysis 4 indicated a consistent bias toward the visual modality when detectors operate in a single-modality scenario, suggesting that late-fusion architectures often lead to modality competition rather than complementarity Future Directions The key takeaway messages suggest that current multimodal detectors rely more on visual artifacts and need new architectures and fusion strategies to mitigate modality bias Future work should explore finetuning the second-stage backbones or randomly masking the dominant modality during training to achieve more balanced results Mitigating artifact bias in visual modalities is key to learning richer and more diverse forgery cues The paper concludes that Mega-MMDF and DeepfakeBench-MM provide a solid foundation for future research in multimodal deepfake detection In this work, the authors first introduce Mega-MMDF, a million-scale, diverse, and high-quality dataset designed to advance research in multimodal deepfake detection Building upon it, they develop DeepfakeBench-MM, the first unified benchmark that standardizes preprocessing and evaluation protocols for multimodal deepfake detection Leveraging these infrastructures allows for fair comparisons and facilitates the development of new methods The release of Mega-MMDF and DeepfakeBench-MM is believed to provide a solid foundation for future research in this area References Enes Altuncu, Virginia NL Franqueira, and Shujun Li.

Nadia: Deepfake: Definitions, performance metrics and standards, datasets, and a meta-review.

Elias: Frontiers in Big Data, 2024 Junyi Ao et al.

Nadia: Speecht5: Unified-modal encoder-decoder pre-training for spoken language processing.

Paper summary: Nadia: It’s about providing the infrastructure needed so we can actually compare different detection methods fairly against one another, which is a big step forward for how we approach defense against deepfakes in practice.

Elias: The authors are addressing the lack of sufficient and diverse training data, and they've built a comprehensive benchmark that tackles both data scarcity and evaluation standardization at once.

Priya: From my side, I think the real value here is moving away from detectors that only work well on one specific type of forgery artifact toward systems that can handle the full spectrum of multimodal fakes we see in the wild.

Nadia: Yeah, so we have this massive dataset and a standardized testbed to push the limits on what multimodal deepfake detection can actually achieve when it has to deal with real-world complexity.

Conclusion: Nadia: So we’ve seen how they built this huge set of real and fake data, and now they have a way to test all these different detectors against it all at once.

Elias: They call it DeepfakeBench-MM, and honestly, the authors are focused on making sure everyone is using the same rules for training and testing these detection models.

Priya: What I see here is that by standardizing the setup, they’re trying to get past just seeing if one detector works well on one dataset.

Nadia: Right, so instead of just saying a model got a high score on Mega-MMDF and then hoping it works elsewhere, they are testing how well it generalizes across different kinds of fakes and even different types of data.

Elias: They’ve included five different deepfake datasets and eleven detectors to see if those models can actually handle the variety.

Priya: That standardization is important because it forces us to look at where those models fail, not just where they succeed on a neat little test set.

Nadia: It really comes down to whether these tools are robust enough for real-world use against the messy stuff people are actually making now.

Elias: The authors have built a solid foundation here, and it gives us a common place to start looking at how well our detection strategies hold up under pressure.

Priya: And looking at the findings, they point out that there’s still this bias towards visual details in some single-modality setups, which tells us we need better ways to balance what the system is actually paying attention to.

Nadia: Yeah, so it suggests that future work needs to focus on making those models less reliant on just the visual stuff when they are dealing with audio too.

Elias: Exactly, and that’s where we go next because understanding this limitation tells us exactly where we need to push our research in the next few years.

The Chinese University of Hong Kong, Shenzhen · University at Buffalo, State University of New York · Nanjing University of Posts and Telecommunications

cs.CR, cs.CV, cs.MM

Submitted: 2025-10-26

Updated: 2026-11-14

Comments: Preprint

Code: https://github.com/haofanwang/inswapper

License: http://creativecommons.org/licenses/by-nc-sa/4.0/

Importance score: 98/100

The gist: DeepfakeBench-MM introduces Mega-MMDF, a large-scale and diverse multimodal deepfake dataset and DeepfakeBench-MM…

Key concepts

Mega-MMDF
A massive, high-quality dataset containing 1.2 million real and forged samples. It was created using an automated framework that generated 21 compositionally diverse forgery pipelines based on various methods and types, ensuring large scale and high diversity.
DeepfakeBench-MM
The first unified benchmark for multimodal deepfake detection. It supports five different multimodal datasets and eleven distinct detectors within an extensible codebase. It standardizes protocols for training and evaluation, allowing for fair comparisons across different methods.
Multimodal Deepfake Detection
The field of detecting manipulated media that combines visual (video) and auditory (audio) elements. This paper focuses on creating tools and benchmarks to accurately identify deepfakes where both the video content and the accompanying sound are intentionally forged or manipulated.
Modality Competition
A phenomenon observed in single-modality detectors where they show a consistent bias toward the visual modality. This suggests that late-fusion architectures might lead to competition between modalities rather than complementary information when detecting deepfakes.

Terminology

Summary

DeepfakeBench-MM introduces Mega-MMDF, a large-scale and diverse multimodal deepfake dataset and DeepfakeBench-MM, the first unified benchmark for multimodal deepfake detection The paper presents this infrastructure to address the lack of sufficient and diverse training data and standardized benchmarks hindering deeper exploration

Dataset Construction

Mega-MMDF is a large-scale multimodal deepfake dataset constructed through an automated deepfake generation framework, featuring three key characteristics: large scale, high forgery diversity, and high quality The dataset contains 0.1 million real samples and 1. p. 0

Where the paper says it

Mega-MMDF currently contains 0. 1 million real samples and 1.

1 million forged samples, totaling 1. p. 0
Where the paper says it

1 million real samples and 1. 1 million forged samples, making it one of the largest and most diverse multimodal deepfake datasets, with plans for continuous expansion.

2 million instances The construction involves generating 21 compositionally diverse forgery pipelines based on combinations of 28 forgery methods across 7 forgery types Quality control is integrated into every pipeline through three assessment modules: audio quality assessment, visual quality assessment, and audio-video synchronization quality assessment Overall dataset quality was validated using objective quantitative metrics like Fréchet Audio Distance (FAD) and Fréchet Video Distance (FVD), alongside human studies involving 60 volunteers

Benchmark Development

DeepfakeBench-MM is the first unified benchmark for multimodal deepfake detection, supporting 5 multimodal deepfake datasets and 11 detectors within an extensible codebase The benchmark standardizes training/evaluation protocols across detection components and provides a common platform for method analysis Supported datasets include FakeAVCeleb, LAV-DF, AVDeepfake1M, IDForge, and Mega-MMDF Detectors implemented span four categories: Baseline Models, Regular Models (7 detectors), Ensemble Models (one model), and Pretrained Multimodal Large Language Models (two models) The evaluation protocol focuses on intra-dataset performance, cross-dataset generalization, and cross-pipeline generalization using the AUC metric

Analysis of Findings

Analysis 1 observed that although detectors generally achieve high AUCs within Mega-MMDF (up to 0.994), their performance drops markedly when applied to external datasets Analysis 2 revealed that in stacked forgeries, artifacts contribute unequally, following the hierarchy EFS > FS > FR Analysis 3 showed that finetuning the second-stage backbones of detectors like AVTS and AVFF yields consistent improvements across datasets Analysis 4 indicated a consistent bias toward the visual modality when detectors operate in a single-modality scenario, suggesting that late-fusion architectures often lead to modality competition rather than complementarity

Future Directions

The key takeaway messages suggest that current multimodal detectors rely more on visual artifacts and need new architectures and fusion strategies to mitigate modality bias Future work should explore finetuning the second-stage backbones or randomly masking the dominant modality during training to achieve more balanced results Mitigating artifact bias in visual modalities is key to learning richer and more diverse forgery cues The paper concludes that Mega-MMDF and DeepfakeBench-MM provide a solid foundation for future research in multimodal deepfake detection

Conclusion

In this work, the authors first introduce Mega-MMDF, a million-scale, diverse, and high-quality dataset designed to advance research in multimodal deepfake detection Building upon it, they develop DeepfakeBench-MM, the first unified benchmark that standardizes preprocessing and evaluation protocols for multimodal deepfake detection Leveraging these infrastructures allows for fair comparisons and facilitates the development of new methods The release of Mega-MMDF and DeepfakeBench-MM is believed to provide a solid foundation for future research in this area

References

Enes Altuncu, Virginia NL Franqueira, and Shujun Li. Deepfake: Definitions, performance metrics and standards, datasets, and a meta-review. p. 9

Where the paper says it

REFERENCES Enes Altuncu, Virginia NL Franqueira, and Shujun Li. Deepfake: Definitions, performance metrics and standards, datasets, and a meta-review.

Frontiers in Big Data, 2024 Junyi Ao et al. p. 9
Where the paper says it

Frontiers in Big Data, 2024. Junyi Ao, Rui Wang, Long Zhou, Chengyi Wang, Shuo Ren, Yu Wu, Shujie Liu, Tom Ko, Qing Li, Yu Zhang, et al.

Speecht5: Unified-modal encoder-decoder pre-training for spoken language processing. p. 9
Where the paper says it

Junyi Ao, Rui Wang, Long Zhou, Chengyi Wang, Shuo Ren, Yu Wu, Shujie Liu, Tom Ko, Qing Li, Yu Zhang, et al. Speecht5: Unified-modal encoder-decoder pre-training for spoken language processing.

In Association for Computational Linguistics, 2022 Max Bain et al. p. 9
Where the paper says it

In Association for Computational Linguistics, 2022. Max Bain, Jaesung Huh, Tengda Han, and Andrew Zisserman.

Whisperx: Time-accurate speech transcription of long-form audio. p. 9
Where the paper says it

Max Bain, Jaesung Huh, Tengda Han, and Andrew Zisserman. Whisperx: Time-accurate speech transcription of long-form audio.

Interspeech, 2023 Zhixi Cai et al. p. 9
Where the paper says it

Interspeech, 2023. Zhixi Cai, Shreya Ghosh, Abhinav Dhall, Tom Gedeon, Kalin Stefanov, and Munawar Hayat.

Glitch in the matrix: A large scale benchmark for content driven audio–visual forgery detection and localization. p. 9
Where the paper says it

Zhixi Cai, Shreya Ghosh, Abhinav Dhall, Tom Gedeon, Kalin Stefanov, and Munawar Hayat. Glitch in the matrix: A large scale benchmark for content driven audio–visual forgery detection and localization.

Computer Vision and Image Understanding, 2023 Zhixi Cai et al. p. 9
Where the paper says it

Computer Vision and Image Understanding, 2023. Zhixi Cai, Shreya Ghosh, Aman Pankaj Adatia, Munawar Hayat, Abhinav Dhall, Tom Gedeon, and Kalin Stefanov.

Av-deepfake1m: A large-scale llm-driven audio-visual deepfake dataset. p. 9
Where the paper says it

Zhixi Cai, Shreya Ghosh, Aman Pankaj Adatia, Munawar Hayat, Abhinav Dhall, Tom Gedeon, and Kalin Stefanov. Av-deepfake1m: A large-scale llm-driven audio-visual deepfake dataset.

In ACM International Conference on Multimedia, 2024 Junyi Cao et al. p. 9
Where the paper says it

In ACM International Conference on Multimedia, 2024. Junyi Cao, Chao Ma, Taiping Yao, Shen Chen, Shouhong Ding, and Xiaokang Yang.

End-to-end reconstruction–classification learning for face forgery detection. p. 9
Where the paper says it

Junyi Cao, Chao Ma, Taiping Yao, Shen Chen, Shouhong Ding, and Xiaokang Yang. End-to-end reconstruction-classification learning for face forgery detection.

In Computer Vision and Pattern Recognition, 2022 Edresson Casanova et al. p. 9
Where the paper says it

In Computer Vision and Pattern Recognition, 2022. Edresson Casanova, Julian Weber, Christopher D Shulby, Arnaldo Candido Junior, Eren Gölge, and Moacir A Ponti.

Yourtts: Towards zero-shot multi-speaker tts and zero-shot voice conversion for everyone. p. 9
Where the paper says it

Edresson Casanova, Julian Weber, Christopher D Shulby, Arnaldo Candido Junior, Eren Gölge, and Moacir A Ponti. Yourtts: Towards zero-shot multi-speaker tts and zero-shot voice conversion for everyone.

In International Conference on Machine Learning, 2022 Abhra Chaudhuri et al. p. 9
Where the paper says it

Yourtts: Towards zero-shot multi-speaker tts and zero-shot voice conversion for everyone. In International Conference on Machine Learning, 2022.

A closer look at multimodal representation collapse. p. 9
Where the paper says it

Abhra Chaudhuri, Anjan Dutta, Tu Bui, and Serban Georgescu. A closer look at multimodal representation collapse.

In International Conference on Machine Learning, 2025 Zesen Cheng et al. p. 9
Where the paper says it

In International Conference on Machine Learning, 2025. Zesen Cheng, Sicong Leng, Hang Zhang, Yifei Xin, Xin Li, Guanzheng Chen, Yongxin Zhu, Wenqi Zhang, Ziyang Luo, Deli Zhao, and Lidong Bing.

Videollama 2: Advancing spatial-temporal modeling and audio understanding in video-llms. p. 9
Where the paper says it

Zesen Cheng, Sicong Leng, Hang Zhang, Yifei Xin, Xin Li, Guanzheng Chen, Yongxin Zhu, Wenqi Zhang, Ziyang Luo, Deli Zhao, and Lidong Bing. Videollama 2: Advancing spatial-temporal modeling and audio understanding in video-llms.

arXiv e-prints, 2024 Ha-Yeong Choi et al. p. 9
Where the paper says it

arXiv e-prints, 2024. Ha-Yeong Choi, Sang-Hoon Lee, and Seong-Whan Lee.

Diff-hiervc: Diffusion-based hierarchical voice conversion with robust pitch generation and masked prior for zero-shot speaker adaptation. p. 9
Where the paper says it

Ha-Yeong Choi, Sang-Hoon Lee, and Seong-Whan Lee. Diff-hiervc: Diffusion-based hierarchical voice conversion with robust pitch generation and masked prior for zero-shot speaker adaptation.

In Interspeech, 2023 Kaiming He et al. p. 9
Where the paper says it

Whisperx: Time-accurate speech transcription of long-form audio. Interspeech, 2023.

Masked autoencoders are scalable vision learners. p. 9
Where the paper says it

Ha-Yeong Choi, Sang-Hoon Lee, and Seong-Whan Lee. Diff-hiervc: Diffusion-based hierarchical voice conversion with robust pitch generation and masked prior for zero-shot speaker adaptation.

In Computer Vision and Pattern Recognition, 2022 Yinan He et al. p. 9
Where the paper says it

End-to-end reconstruction-classification learning for face forgery detection. In Computer Vision and Pattern Recognition, 2022.

Forgerynet: A versatile benchmark for comprehensive forgery analysis. p. 0
Where the paper says it

Building on it, we present DeepfakeBench-MM, the first unified benchmark for multimodal deepfake detection. It establishes standardized protocols across the entire detection pipeline and serves as a versatile platform for evaluating existing methods as well as exploring novel approaches.

In Computer Vision and Pattern Recognition, 2021 Sindhu Hegde et al. p. 9
Where the paper says it

End-to-end reconstruction-classification learning for face forgery detection. In Computer Vision and Pattern Recognition, 2022.

Towards accurate lip-to-speech synthesis in-the-wild. p. 9
Where the paper says it

Max Bain, Jaesung Huh, Tengda Han, and Andrew Zisserman. Whisperx: Time-accurate speech transcription of long-form audio.

In ACM International Conference on Multimedia, 2023 Jie Hu et al. p. 9
Where the paper says it

Av-deepfake1m: A large-scale llm-driven audio-visual deepfake dataset. In ACM International Conference on Multimedia, 2024.

Squeeze-and-excitation networks. In Computer Vision and Pattern Recognition, 2018 Yu Huang et al. p. 9
Where the paper says it

End-to-end reconstruction-classification learning for face forgery detection. In Computer Vision and Pattern Recognition, 2022.

Modality competition: What makes joint training of multi-modal network fail in deep learning? p. 8
Where the paper says it

, 2025), in which multimodal models fail to leverage both modalities in a balanced manner. Since most detectors adopt a late-fusion architecture, the two modalities often act in competition rather than complementarity.

(provably). In International conference on machine learning, 2022 Ziyao Huang et al. p. 9
Where the paper says it

Yourtts: Towards zero-shot multi-speaker tts and zero-shot voice conversion for everyone. In International Conference on Machine Learning, 2022.

Identity-preserving face swapping via dual surrogate generative models. p. 3
Where the paper says it

, 2024), we consider three common types of visual-only forgeries: Face Swapping (FS), Entire Face Synthesis (EFS), and Face Editing (FE). FS swaps the facial ID of the target video with the ID of the source video, while preserving all remaining contents (e.

ACM Transactions on Graphics, 2024 Liming Jiang et al. p. 9
Where the paper says it

Av-deepfake1m: A large-scale llm-driven audio-visual deepfake dataset. In ACM International Conference on Multimedia, 2024.

Deeperforensics-1.0: A largescale dataset for real-world face forgery detection. p. 3
Where the paper says it

For visual forgery methods that output static images, we employ audio-driven face reenactment (FR) to generate temporally consistent videos that align with the audio. Similarly, for video-based visual forgeries where the manipulated video does not naturally synchronize with the forged audio, we apply face reenactment to enforce cross-modal consistency, mimicking real-world deepfake scenarios.

In Computer Vision and Pattern Recognition, 2020 Haofan Wang. p. 9
Where the paper says it

End-to-end reconstruction-classification learning for face forgery detection. In Computer Vision and Pattern Recognition, 2022.

Inswapper. https://github.com/haofanwang/inswapper, 2023 Kaisiyuan Wang et al. p. 3
Where the paper says it

1 REAL DATA COLLECTION Our real-world audiovisual samples are drawn from three public datasets: Voxceleb2 (Chung et al. , 2018), MEAD (Wang et al.

Mead: A large-scale audio-visual dataset for emotional talking-face generation. p. 0
Where the paper says it

To address this challenge, we first build Mega-MMDF, a large-scale, diverse, and high-quality dataset for multimodal deepfake detection. Specifically, we employ 21 forgery pipelines through the combination of 10 audio forgery methods, 12 visual forgery methods, and 6 audio-driven face reenactment methods.

In European Conference on Computer Vision, 2020a Philip Wang. p. 3
Where the paper says it

, 2018), MEAD (Wang et al. , 2020a), and CelebV-HQ (Zhu et al.

This person does not exist. https://thispersonnotexist.org//, 2019 Xin Wang et al. p. 4
Where the paper says it

4 OVERALL QUALITY After the dataset construction process, we validate overall dataset quality through (1) Objective Quantitative Metrics using Fréchet Audio Distance (FAD) (Kilgour et al. , 2019) and Fréchet Video Distance (FVD) (Unterthiner et al.

Asvspoof 2019: A large-scale public database of synthesized, converted and replayed speech. Computer Speech & Language, 2020b Zhenbang Wu et al. p. 3
Where the paper says it

, 2020b; Wu et al. , 2015), we consider three common types of audio forgeries: Text-to-Speech (TTS), Voice Conversion (VC), and Partial Fake (PF).

Multimodal patient representation learning with missing modalities and labels. p. 8
Where the paper says it

This architecture can lead to predictive cues from one modality to become entangled with noise features from the other, suppressing informative signals (Chaudhuri et al. While such modality bias has been extensively discussed in multimodal representation learning, it remains largely unaddressed in current multimodal deepfake detection frameworks, highlighting a critical design limitation that suggests future detectors should incorporate strategies, such as bias-aware training or adaptive fusion, to mitigate modality dominance and fully exploit the strengths of each modality.

In International Conference on Learning Representations, 2024 Zhizheng Wu et al. p. 9
Where the paper says it

Yourtts: Towards zero-shot multi-speaker tts and zero-shot voice conversion for everyone. In International Conference on Machine Learning, 2022.

Asvspoof 2015: the first automatic speaker verification spoofing and countermeasures challenge. p. 3
Where the paper says it

, 2015), we consider three common types of audio forgeries: Text-to-Speech (TTS), Voice Conversion (VC), and Partial Fake (PF). Given a text input and reference audio, TTS generates fake audio adhering to the speaker’s tone of the reference audio.

In Interspeech, 2015 Chao Xu et al. p. 3
Where the paper says it

, 2020b; Wu et al. , 2015), we consider three common types of audio forgeries: Text-to-Speech (TTS), Voice Conversion (VC), and Partial Fake (PF).

Designing one unified framework for high-fidelity face reenactment and swapping. p. 0
Where the paper says it

To address this challenge, we first build Mega-MMDF, a large-scale, diverse, and high-quality dataset for multimodal deepfake detection. Specifically, we employ 21 forgery pipelines through the combination of 10 audio forgery methods, 12 visual forgery methods, and 6 audio-driven face reenactment methods.

In European Conference on Computer Vision, 2022 Jin Xu et al. p. 9
Where the paper says it

End-to-end reconstruction-classification learning for face forgery detection. In Computer Vision and Pattern Recognition, 2022.

Qwen2.5-omni technical report. p. 5
Where the paper says it

For consistency with prior work (Yan et al. , 2023; 2024), we report results using the widely adopted AUC metric.

arXiv e-prints, 2025 Junhao Xu et al. p. 9
Where the paper says it

Videollama 2: Advancing spatial-temporal modeling and audio understanding in video-llms. arXiv e-prints, 2024.

Identity-driven multimedia forgery detection via reference assistance. p. 9
Where the paper says it

Av-deepfake1m: A large-scale llm-driven audio-visual deepfake dataset. In ACM International Conference on Multimedia, 2024.

In ACM International Conference on Multimedia, 2024 Yuezun Li et al. p. 9
Where the paper says it

Av-deepfake1m: A large-scale llm-driven audio-visual deepfake dataset. In ACM International Conference on Multimedia, 2024.

Celeb-df: A large-scale challenging dataset for deepfake forensics. p. 0
Where the paper says it

However, the lack of sufficient and diverse training data, along with the absence of a standardized benchmark, hinder deeper exploration. To address this challenge, we first build Mega-MMDF, a large-scale, diverse, and high-quality dataset for multimodal deepfake detection.

In Computer Vision and Pattern Recognition, 2020 Zhen Li et al. p. 9
Where the paper says it

End-to-end reconstruction-classification learning for face forgery detection. In Computer Vision and Pattern Recognition, 2022.

Photomaker: Customizing realistic human photos via stacked id embedding. p. 4
Where the paper says it

, 2018) to measure audio/video fidelity by quantifying real-forgery sample distribution gaps, and to evaluate synchronization using Sync-C and Sync-D calculated via a pretrained SyncNet (Chung & Zisserman, 2017) model. (2) Human Studies involving 60 volunteers from diverse backgrounds (e.

In Computer Vision and Pattern Recognition, 2024b Wenxuan Zhang et al. p. 9
Where the paper says it

End-to-end reconstruction-classification learning for face forgery detection. In Computer Vision and Pattern Recognition, 2022.

Sadtalker: Learning realistic 3d motion coefficients for stylized audio-driven single image talking face animation. p. 2
Where the paper says it

, 2024; Altuncu et al. , 2024) have provided comprehensive overviews of single-modality datasets, spanning image-, audio-, and video-based forgeries.

In Computer Vision and Pattern Recognition, 2023 Xueyao Zhang et al. p. 9
Where the paper says it

End-to-end reconstruction-classification learning for face forgery detection. In Computer Vision and Pattern Recognition, 2022.

Vevo: Controllable zero-shot voice imitation with self-supervised disentanglement. p. 9
Where the paper says it

Edresson Casanova, Julian Weber, Christopher D Shulby, Arnaldo Candido Junior, Eren Gölge, and Moacir A Ponti. Yourtts: Towards zero-shot multi-speaker tts and zero-shot voice conversion for everyone.

In International Conference on Learning Representations, 2025 Hao Zhu et al. p. 9
Where the paper says it

In International Conference on Machine Learning, 2025. Zesen Cheng, Sicong Leng, Hang Zhang, Yifei Xin, Xin Li, Guanzheng Chen, Yongxin Zhu, Wenqi Zhang, Ziyang Luo, Deli Zhao, and Lidong Bing.

Celebv-hq: A large-scale video facial attributes dataset. p. 7
Where the paper says it

Several multimodal deepfake detectors, such as AVFF and AVTS, adopt a two-phase paradigm. Backbones are first pretrained on large-scale real audio-video pairs to learn cross-modal synchronization, and then trained on deepfake datasets for classification.

p. 7
Where the paper says it

Several multimodal deepfake detectors, such as AVFF and AVTS, adopt a two-phase paradigm. Backbones are first pretrained on large-scale real audio-video pairs to learn cross-modal synchronization, and then trained on deepfake datasets for classification.

In European Conference on Computer Vision, 2022 Chang-Sung Sung et al. p. 9
Where the paper says it

End-to-end reconstruction-classification learning for face forgery detection. In Computer Vision and Pattern Recognition, 2022.

Hearing and seeing abnormality: Selfsupervised audio-visual mutual learning for deepfake detection. p. 0
Where the paper says it

To address this challenge, we first build Mega-MMDF, a large-scale, diverse, and high-quality dataset for multimodal deepfake detection. Specifically, we employ 21 forgery pipelines through the combination of 10 audio forgery methods, 12 visual forgery methods, and 6 audio-driven face reenactment methods.

In International Conference on Acoustics, Speech and Signal Processing, 2023 Shuai Tan et al. p. 4
Where the paper says it

, 2023), a speech-to-text (STT) model. (b) SNR quantifies the ratio of meaningful signal to background noise.

Edtalk: Efficient disentanglement for emotional talking head synthesis. p. 3
Where the paper says it

, 2024; 2023; Cai et al. , 2024), we consider three common types of visual-only forgeries: Face Swapping (FS), Entire Face Synthesis (EFS), and Face Editing (FE).

In European Conference on Computer Vision, 2024 Kartik Thakral et al. p. 9
Where the paper says it

Glitch in the matrix: A large scale benchmark for content driven audio–visual forgery detection and localization. Computer Vision and Image Understanding, 2023.

ILLUSION: Unveiling truth with a comprehensive multi-modal, multi-lingual deepfake dataset. p. 7
Where the paper says it

The dominance of EFS in certain combinations reflects an artifact bias phenomenon, where specific artifacts disproportionately influence the learned representation. This bias can skew cross-pipeline generalization and highlights the need for future work on representation balancing to ensure fairer exploitation of multi-artifact information.

In International Conference on Learning Representations, 2025 Du Tran et al. p. 9
Where the paper says it

A closer look at multimodal representation collapse. In International Conference on Machine Learning, 2025.

Learning spatiotemporal features with 3d convolutional networks. p. 8
Where the paper says it

This architecture can lead to predictive cues from one modality to become entangled with noise features from the other, suppressing informative signals (Chaudhuri et al. While such modality bias has been extensively discussed in multimodal representation learning, it remains largely unaddressed in current multimodal deepfake detection frameworks, highlighting a critical design limitation that suggests future detectors should incorporate strategies, such as bias-aware training or adaptive fusion, to mitigate modality dominance and fully exploit the strengths of each modality.

In International Conference on Computer Vision, 2015 Du Tran et al. p. 9
Where the paper says it

Glitch in the matrix: A large scale benchmark for content driven audio–visual forgery detection and localization. Computer Vision and Image Understanding, 2023.

Convnet architecture search for spatiotemporal feature learning. p. 8
Where the paper says it

This architecture can lead to predictive cues from one modality to become entangled with noise features from the other, suppressing informative signals (Chaudhuri et al. While such modality bias has been extensively discussed in multimodal representation learning, it remains largely unaddressed in current multimodal deepfake detection frameworks, highlighting a critical design limitation that suggests future detectors should incorporate strategies, such as bias-aware training or adaptive fusion, to mitigate modality dominance and fully exploit the strengths of each modality.

arXiv e-prints, 2017 Thomas Unterthiner et al. p. 9
Where the paper says it

Videollama 2: Advancing spatial-temporal modeling and audio understanding in video-llms. arXiv e-prints, 2024.

Towards accurate generative models of video: A new metric & challenges. p. 0
Where the paper says it

Preprint DEEPFAKEBENCH-MM: A COMPREHENSIVE BENCHMARK FOR MULTIMODAL DEEPFAKE DETECTION Kangran Zhao1∗ , Yupeng Chen1∗ , Xiaoyu Zhang1∗ , Yize Chen1 , Weinan Guan1 Baicheng Chen1 , Chengzhe Sun2 , Soumyya Kanti Datta2 Qingshan Liu3 , Siwei Lyu2 , Baoyuan Wu1† 1The Chinese University of Hong Kong, Shenzhen 2University at Buffalo, State University of New York 3Nanjing University of Posts and Telecommunications ABSTRACT The misuse of advanced generative AI models has resulted in the widespread proliferation of falsified data, particularly forged human-centric audiovisual content, which poses substantial societal risks (e. , financial fraud and social instability).

arXiv e-prints, 2018 Laurens Van der Maaten and Geoffrey Hinton. p. 9
Where the paper says it

Videollama 2: Advancing spatial-temporal modeling and audio understanding in video-llms. arXiv e-prints, 2024.

Visualizing data using t-sne. p. 1
Where the paper says it

, 2024), which contains 860,039 fake samples, the scale of fake data in other datasets remains relatively small. (2) Low forgery diversity: Most datasets are generated using only a few forgery methods (e.

Journal of Machine Learning Research, 2008 Haofan Wang. p. 9
Where the paper says it

Yourtts: Towards zero-shot multi-speaker tts and zero-shot voice conversion for everyone. In International Conference on Machine Learning, 2022.

Inswapper. https://github.com/haofanwang/inswapper, 2023 Kaede Shiohara et al. p. 0
Where the paper says it

, 2024; 2022; Li et al. , 2023), facial images/videos (Shiohara et al.

Blendface: Re-designing identity encoders for face-swapping. p. 3
Where the paper says it

, 2024; 2023; Cai et al. , 2024), we consider three common types of visual-only forgeries: Face Swapping (FS), Entire Face Synthesis (EFS), and Face Editing (FE).

In International Conference on Computer Vision, 2023 Yuchen Luo et al. p. 9
Where the paper says it

In International Conference on Machine Learning, 2025. Zesen Cheng, Sicong Leng, Hang Zhang, Yifei Xin, Xin Li, Guanzheng Chen, Yongxin Zhu, Wenqi Zhang, Ziyang Luo, Deli Zhao, and Lidong Bing.

Generalizing face forgery detection with high-frequency features. p. 2
Where the paper says it

FRADE (Nie et al. , 2024) introduces an AFI adapter to improve attention to high-frequency details and an ACI module to fuse audio and visual features.

In Computer Vision and Pattern Recognition, 2021 Anish Mittal et al. p. 9
Where the paper says it

End-to-end reconstruction-classification learning for face forgery detection. In Computer Vision and Pattern Recognition, 2022.

No-reference image quality assessment in the spatial domain. p. 3
Where the paper says it

Details of the 28 forgery methods and the 21 forgery pipelines are provided in Appendix A. 3 QUALITY CONTROL MECHANISM DURING DEEPFAKE DATA CONSTRUCTION To ensure the quality of the constructed forgery data, we design a quality control mechanism comprising three assessment modules: audio quality assessment, visual quality assessment, and audio-video synchronization quality assessment.

IEEE Transactions on Image Processing, 2012 Soumik Mukhopadhyay et al. Diff2lip: Audio conditioned diffusion models for lip-synchronization. p. 4
Where the paper says it

, 2024) to assess the lip synchronization quality of manipulated audiovisual data. 4 OVERALL QUALITY After the dataset construction process, we validate overall dataset quality through (1) Objective Quantitative Metrics using Fréchet Audio Distance (FAD) (Kilgour et al.

In Winter Conference on Applications of Computer Vision, 2024 Kartik Narayan et al. p. 9
Where the paper says it

Glitch in the matrix: A large scale benchmark for content driven audio–visual forgery detection and localization. Computer Vision and Image Understanding, 2023.

Dfplatter: Multi-face heterogeneous deepfake dataset. p. 0
Where the paper says it

To address this challenge, we first build Mega-MMDF, a large-scale, diverse, and high-quality dataset for multimodal deepfake detection. Specifically, we employ 21 forgery pipelines through the combination of 10 audio forgery methods, 12 visual forgery methods, and 6 audio-driven face reenactment methods.

In Computer Vision and Pattern Recognition, 2023 Fan Nie et al. p. 9
Where the paper says it

End-to-end reconstruction-classification learning for face forgery detection. In Computer Vision and Pattern Recognition, 2022.

Frade: Forgery-aware audiodistilled multimodal learning for deepfake detection. p. 8
Where the paper says it

This architecture can lead to predictive cues from one modality to become entangled with noise features from the other, suppressing informative signals (Chaudhuri et al. While such modality bias has been extensively discussed in multimodal representation learning, it remains largely unaddressed in current multimodal deepfake detection frameworks, highlighting a critical design limitation that suggests future detectors should incorporate strategies, such as bias-aware training or adaptive fusion, to mitigate modality dominance and fully exploit the strengths of each modality.

In ACM International Conference on Multimedia, 2024 Kwon et al. p. 9
Where the paper says it

Av-deepfake1m: A large-scale llm-driven audio-visual deepfake dataset. In ACM International Conference on Multimedia, 2024.

Kodf: A large-scale korean deepfake detection dataset. p. 0
Where the paper says it

However, the lack of sufficient and diverse training data, along with the absence of a standardized benchmark, hinder deeper exploration. To address this challenge, we first build Mega-MMDF, a large-scale, diverse, and high-quality dataset for multimodal deepfake detection.

In International Conference on Computer Vision, 2021 Seonghak Lee et al. p. 9
Where the paper says it

Glitch in the matrix: A large scale benchmark for content driven audio–visual forgery detection and localization. Computer Vision and Image Understanding, 2023.

Wildtalker: Talking portrait synthesis in the wild. p. 1
Where the paper says it

(2) Low forgery diversity: Most datasets are generated using only a few forgery methods (e. , AVDeepfake1M uses three while FakeAVCeleb uses four), making detectors prone to overfitting to specific forgery artifacts and failing to generalize to the wide variety of forgeries encountered in the wild.

In European Conference on Computer Vision, 2024 Chunyu Li et al. p. 9
Where the paper says it

Glitch in the matrix: A large scale benchmark for content driven audio–visual forgery detection and localization. Computer Vision and Image Understanding, 2023.

Latentsync: Audio conditioned latent diffusion models for lip sync. p. 6
Where the paper says it

By contrast, AVH benefits from pretrained lip-reading checkpoints, enabling it to exploit cross-modal consistency and achieve superior performance in detecting partial manipulations. Moreover, we note that pretrained large models, when evaluated in a zero-shot setting, obtain the lowest AUC scores on Mega-MMDF, highlighting both the difficulty and the novelty of our dataset.

arXiv e-prints, 2024a Jingyi Li et al. p. 9
Where the paper says it

Videollama 2: Advancing spatial-temporal modeling and audio understanding in video-llms. arXiv e-prints, 2024.

Freevc: Towards high-quality text-free one-shot voice conversion. p. 9
Where the paper says it

Edresson Casanova, Julian Weber, Christopher D Shulby, Arnaldo Candido Junior, Eren Gölge, and Moacir A Ponti. Yourtts: Towards zero-shot multi-speaker tts and zero-shot voice conversion for everyone.

In International Conference on Acoustics, Speech and Signal Processing, 2023 Yue Zhang et al. p. 9
Where the paper says it

In International Conference on Machine Learning, 2025. Zesen Cheng, Sicong Leng, Hang Zhang, Yifei Xin, Xin Li, Guanzheng Chen, Yongxin Zhu, Wenqi Zhang, Ziyang Luo, Deli Zhao, and Lidong Bing.

Musetalk: Real-time high quality lip synchronization with latent space inpainting. p. 4
Where the paper says it

, 2023; Lee et al. , 2024) to assess the lip synchronization quality of manipulated audiovisual data.

arXiv e-prints, 2024 Hao Zhu et al. p. 9
Where the paper says it

Videollama 2: Advancing spatial-temporal modeling and audio understanding in video-llms. arXiv e-prints, 2024.

Celebv-hq: A large-scale video facial attributes dataset. In European Conference on Computer Vision, 2022 Heqing Zou et al. p. 9
Where the paper says it

End-to-end reconstruction-classification learning for face forgery detection. In Computer Vision and Pattern Recognition, 2022.

Crossmodality and within-modality regularization for audio-visual deepfake detection. p. 8
Where the paper says it

This approach can be regarded as a bias-mitigation form of data augmentation, which is rarely explored in multimodal forgery detection. Given that models trained on Mega-MMDF generally exhibit a strong preference for the visual modality, we randomly mask the video stream during training and supervise using audio-modality labels.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Augment multimodal deepfake detectors with finetuning of second-stage backbones to adapt to forgery-specific artifacts, as shown by finetuning yields consistent improvements across datasets in Table 3. This allows models like AVTS and AVFF to overcome the limitations of frozen backbones by adapting backbone representations to artifact-specific cues.

  2. Implement a training strategy involving random modality masking during training to encourage better exploitation of weaker modalities, as suggested by randomly masking the video stream during training and supervise using audio-modality labels for AVTS and MRDF. This aims to mitigate modality dominance in models that show a consistent bias toward the visual modality.

  3. Design fusion strategies that explicitly address modality competition rather than simple averaging, aiming to mitigate the modalities often act in competition rather than complementarity issue, as highlighted by the analysis of late-fusion architectures. This could involve developing adaptive fusion methods to ensure complementary use of all modalities.

  4. Develop a hierarchical or artifact-aware representation learning approach that accounts for the hierarchy of forgery artifacts, such as EFS > FS > FR, to improve cross-pipeline generalization across diverse manipulations. This addresses the finding that EFS+FR generalizes much better to other EFS-related pipelines than FS-only pipelines.

  5. Implement a standardized preprocessing pipeline within the benchmark codebase that incorporates hybrid face detection strategies, specifically discarding frames with no detected face or multiple faces, to ensure clear speaker-face correspondence. This standardizes data preparation across heterogeneous datasets like Mega-MMDF and others, preventing issues like background region cropped sometimes.

  6. Utilize pretrained multimodal large language models like Qwen2.5-Omni or VideoLLaMA2 for zero-shot detection by leveraging their robust cross-modal representation and generation capabilities. This allows for immediate evaluation of zero-shot performance on unseen deepfake content without requiring extensive fine-tuning on specific forgery types.

  7. Create an ensemble model that explicitly combines the results from visual and audio detectors using modality-specific backbones, ensuring that each detector focuses on its modality's strengths before combining predictions with an averaging strategy. This serves as a competitive baseline by exploiting complementary information across modalities.

Abstract

Audio-visual deepfakes can spread misinformation and enable impersonation, undermining trust in digital media. However, existing datasets provide limited coverage of forgery operations and their combinations, constraining detector development and systematic evaluation. We introduce Mega-MMDF, containing 100,000 real and 1.2 million forged clips across 22 pipelines spanning modality-specific manipulations, stacked forgeries, and joint audio-video generation. To assess its training value and enable fair comparisons, we develop DeepfakeBench-MM, which standardizes training and evaluation across 8 datasets and 10 detectors. Training-based detectors achieve their highest mean AUC when trained on Mega-MMDF, with gains of up to 0.0784 over the best alternative training dataset for each detector. Experiments on data scale and pipeline coverage further support the value of large, diverse training data. Cross-pipeline evaluations reveal bidirectional and asymmetric generalization. Training on composite forgeries does not necessarily enable detection of their constituent operations in isolation. Our analysis of altered inputs shows that modality masking can increase false alarms and missed detections, and temporal misalignment can raise mean predicted forgery probabilities. Together, these resources provide a foundation for understanding and improving multimodal deepfake detection, with publicly available code and dataset access subject to application and usage terms.

Related papers